OVHcloud Assistant
Ask me anything about OVHcloud !
Introducing our OVHcloud Assistant, now available in ALPHA version. OVHcloud Assistant is designed to provide you with quick and accurate answers to your questions about our services, and to enhance your overall customer experience.
How does it work ?
OVHcloud Assistant leverages a range of cutting-edge technologies to deliver top-notch responses. Here's a breakdown of the key steps involved:
1- OVHcloud page indexing
We use RAG and LLM (LLama 3.3 70B) technologies to index our entire website on a weekly basis. This process generates a vast database of 100,000 pages and 1.2 million lines, which we use as a reference point to inform our chatbot's responses.
2- Search & Reformulation
We use a combination of RAG search and LLM-powered question reformulation to surface the most relevant answers. To further enhance the response, we also leverage Llama 4 17B (with a latency of 800-900 ms) to reformulate the question and provide a snippet of the answer, while DeepSeek 3.1 (with a latency of 2.5-3.5 seconds) is used to digest the RAG results.
3- Response Scoring
We employ the Mistral-medium model to evaluate the accuracy and relevance of our chatbot's responses, and to identify areas where our website content may be lacking. This allows us to pinpoint specific pages or texts that need to be added or updated to provide more comprehensive and accurate information to our users.
OVHcloud Assistant does not have memory or the ability to recall past interactions.
Each conversation is processed independently without storing or tracking previous questions or user information.
Under the hood
Key Technologies :
RAG, LLM
AI Endpoints: LLama 3.3 70B, DeepSeek 3.1, Llama 4 17B, and Mistral-medium
Managed PostgresSQL: 100,000 pages and 1.2 million lines of data
Performance metrics: 800-900 ms for question reformulation and 2.5-3.5 seconds for RAG digestion
Help us improve
Share your feedback
We're looking forward to receiving your feedback and suggestions. This will help us refine our tool and make it even more effective.
-
Alpha
-
Beta
-
General Availability